Status of the QCDSP project

Chen, Dong; Christ, Norman H.; Edwards, Robert G.; Fleming, George; Gara, Alan; Hansen, S.; Jung, Chulwoo; Kaehler, Adrian; Kennedy, A.D.; Kilcup, Gregory W.; Luo, Yubing; Malureanu, C.; Mawhinney, Robert D.; Parsons, J. A.; Sui, C.; Vranas, P.; Zhestkov, Y.

doi:10.1016/s0920-5632(99)85238-4

Cited by 14 publications

(3 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…(Notice that our measured performance is based on double precision code, whereas single precision numbers are estimated ones.) Our PMS1 machine, similarly to other workstation farms has a moderate maximum sustained performance as compared to Teraflop-scale machines (CP-PACS [6] or QCDSP [7]). However, PMS1 has a much better price/performance ratio than other workstation farms.…”

Section: Performancementioning

confidence: 96%

The PMS project: Poor man's supercomputer

Csikor

Fodor

Hegedűs

et al. 2001

Computer Physics Communications

View full text Add to dashboard Cite

We briefly describe the Poor Man's Supercomputer (PMS) project carried out at Eötvös University, Budapest. The goal was to construct a cost effective, scalable, fast parallel computer to perform numerical calculations of physical problems that can be implemented on a lattice with nearest neighbour interactions. To this end we developed the PMS architecture using PC components and designed a special, low cost communication hardware and the driver software for Linux OS. Our first implementation of PMS includes 32 nodes (PMS1). The performance of PMS1 was tested by Lattice Gauge Theory simulations. Using pure SU(3) gauge theory or the bosonic MSSM on PMS1 we obtained 3$/Mflop and 0.45$Mflop price-to-sustained performance ratio for double and single precision operations, respectively. The design of the special hardware and the communication driver are freely available upon request for non-profit organizations. IntroductionOur purpose was to build a high performance supercomputer from PC elements. We use PCs for two reasons. They have excellent cost/performance ratios [1] and can easily be upgraded when faster motherboards and CPUs will be available. The PMS project started in 1998, and the machine is now ready for physical calculations. Our first PMS machine (PMS1) consists of 32 PCs arranged in a three-dimensional 2 × 4 × 4 mesh. Each node has two special communication cards providing fast communication through flat cables to the six neighbours. This gives a much better performance than simple Ethernet link.Since the machine is built from PCs, the latest versions of all programming languages (such as C and Fortran) can be used for coding. Writing applications is straightforward. One only has to keep in mind the 3-dimensional mesh structure of the machine; no further deep understanding of how the communication works is required. There are some routines written in C that make communication of data between the nodes easy. The machine works in Single Instruction Multiple Data (SIMD) mode: all processors execute the same program, while the data they work on may differ.Nowadays double precision floating point arithmetic is necessary for accurate results. PMS offers this precision since the processors have double precision Floating Point Units (FPUs). In cases when single precision is enough, the special MMX instruction set of AMD K6-2 processors can be used. This provides a much higher performance, in principle 8 times higher, than the standard double precision mode.The following sections describe the hardware and software architectures of PMS. We first give a short overview of the machine and then describe the hardware and the software in more detail. Some performance results are also presented, and an outlook is given.

show abstract

Section: Performancementioning

confidence: 96%

The PMS project: Poor man's supercomputer

Csikor

Fodor

Hegedűs

et al. 2001

Computer Physics Communications

View full text Add to dashboard Cite

show abstract

“…Lattice-QCD practitioners have been at the forefront of computing, to the extent of frequently designing special purpose systems that were tuned to the needs of QCD calculations. Notable examples of custom lattice-QCD systems in the U.S. include the ACPMAPS [100], QCDSP [101], and QCDOC [19] supercomputers. ACPMAPS was built from Weitek XL and Intel i860 processors and a custom-built communications backplane such that processing speed, latency, and data bandwith were all well balanced.…”

Section: A Historical Perspective and Machine Buildingmentioning

confidence: 99%

Status and future perspectives for lattice gauge theory calculations to the exascale and beyond

et al. 2019

Self Cite

View full text Add to dashboard Cite

In this and a set of companion whitepapers, the USQCD Collaboration lays out a program of science and computing for lattice gauge theory. These whitepapers describe how calculation using lattice QCD (and other gauge theories) can aid the interpretation of ongoing and upcoming experiments in particle and nuclear physics, as well as inspire new ones. * Editor, bjoo@jlab.org † Editor, chulwoo@bnl.gov arXiv:1904.09725v2 [hep-lat]

show abstract

“…to design teraflop/s-scale computers suitable for computations such as quantum chromodynaniics (QCD). The goal of this collaboration was to develop a machine capable of a peak perforniance of 0.8 Teraflops/s and a true sustained performance of around 35% for QCD [54,55]. DOE has funded a program to build a machine capable of a peak performance of 0.4 Teraflops and a true sustained performance of around 35% for QCD.…”

Section: Computer Architecture and Designmentioning

confidence: 99%

Final report to the DOE for the period 8/1/96 to 5/31/00 by the SCRI Theoretical High Energy Group

Heller¹

2000

View full text Add to dashboard Cite

f6is rrport was prepared as an account of work sponsored by an agency of the U n i d States Government Neither the United States Govcramcat nor any agency thereof, nor any of their employtes makes any warranty, exprrss or implied, or-is ahy kgal liability or responsibility for the accuracy, complercnen, 01 userims of MY information, apparatus. product, or procm d i or rtprcsea!~ that its use wuld not infringe privately owned rights. Rcferrnct her& to any specific cornmadat product, process, or &a by trade name, trademark maaufactorex, or otherwise docs not neatsady constitute or imply its endonemmt. reammead;rtion, or favoring by the Uoitcd S?ts Govcrnmcnt or any agency thmof. The *itan and opinions of authors e x w e d he& do not n d y state or refkct those of the United States GovffnmeDt or my agency thenof. DISCLA! MER Portions of this document may be illegible in electronic image products. Images are produced from the best available original document.

show abstract

Status of the QCDSP project

Cited by 14 publications

References 6 publications

The PMS project: Poor man's supercomputer

The PMS project: Poor man's supercomputer

Status and future perspectives for lattice gauge theory calculations to the exascale and beyond

Final report to the DOE for the period 8/1/96 to 5/31/00 by the SCRI Theoretical High Energy Group

Contact Info

Product

Resources

About